Ammar Amjad
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Ammar Amjad has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Hsien-Tsung Chang
Bachelor Program in Artificial Intelligence, Chang Gung University, Taoyuan, Taiwan
4 shared publications - Lal Khan
Department of AI and SW, Gachon University, Seongnam, Republic of Korea
3 shared publications - Kanwar Muhammad Afaq
Department of Computer Science and Information Engineering, Chang Gung University, Taoyuan, Taiwan
1 shared publication Li-Chia Tai
Department of Electrical and Computer Engineering, National Yang Ming Chiao Tung University, Hsinchu, Taiwan
1 shared publicationMuhammad Bilal Mahmood
Department of Software Engineering, Dalian University of Technology, Dalian, China
1 shared publication- Noman Ashraf
Dana-Farber Cancer Institute, Harvard Medical School, United States
1 shared publication
Latest Publications
- Open Access
- 23 March 2026
- Published by Elsevier in SSRN Electronic Journal
Remote-Sensing-Oriented Diffusion-Based Synthetic Smoke Generation for Early Wildfire Monitoring
Smoke is typically the earliest visible indicator of a wildfire. With the rapid advances in deep learning, image-based smoke detection has become a vital approach for early wildfire detection and prevention. However, a key bottleneck in training effective detection models is the limited availability of annotated forest fire smoke images. While generative models offer a promising solution for synthesizing such data, conventional inpainting approaches often struggle to maintain visual fidelity and contextual consistency, particularly at the interface between synthesized smoke and complex natural backgrounds.To address these challenges, we propose \textbf{FireSmokeGenNet}, which is a new framework for generating realistic and diverse forest fire smoke images. Our pipeline begins by using a pre-trained segmentation model and multimodal vision-language model to extract smoke masks and generate descriptive captions. We then introduce a dual-branch architecture that incorporates both the mask and its masked image to guide the generative process. A key contribution of our framework is the \textit{Mask Random Difference Loss (MRDL)}, which enforces local consistency by introducing stochastic mask perturbations that improve the boundary smoothness and transparency during training.To ensure high-quality synthetic data, we applied a multimodal filtering module powered by a large vision-language model to select contextually valid and perceptually coherent samples. The experimental results demonstrate that images generated by \textit{FireSmokeGenNet} are photorealistic and diverse, significantly improving the performance of downstream forest fire smoke detection tasks. This work introduces a boundary-aware diffusion inpainting framework and a scalable synthetic data pipeline for real-world wildfire monitoring.
- 23 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessRemote-Sensing-Oriented Diffusion-Based Synthetic Smoke Generation for Early Wildfire Monitoring
Smoke is typically the earliest visible indicator of a wildfire. With the rapid advances in deep learning, image-based smoke detection has become a vital approach for early wildfire detection and prevention. However, a key bottleneck in training effective detection models is the limited availability of annotated forest fire smoke images. While generative models offer a promising solution for synthesizing such data, conventional inpainting approaches often struggle to maintain visual fidelity and contextual consistency, particularly at the interface between synthesized smoke and complex natural backgrounds.To address these challenges, we propose \textbf{FireSmokeGenNet}, which is a new framework for generating realistic and diverse forest fire smoke images. Our pipeline begins by using a pre-trained segmentation model and multimodal vision-language model to extract smoke masks and generate descriptive captions. We then introduce a dual-branch architecture that incorporates both the mask and its masked image to guide the generative process. A key contribution of our framework is the \textit{Mask Random Difference Loss (MRDL)}, which enforces local consistency by introducing stochastic mask perturbations that improve the boundary smoothness and transparency during training.To ensure high-quality synthetic data, we applied a multimodal filtering module powered by a large vision-language model to select contextually valid and perceptually coherent samples. The experimental results demonstrate that images generated by \textit{FireSmokeGenNet} are photorealistic and diverse, significantly improving the performance of downstream forest fire smoke detection tasks. This work introduces a boundary-aware diffusion inpainting framework and a scalable synthetic data pipeline for real-world wildfire monitoring.
- Open Access
- 4 June 2025
- Published by Qeios Ltd
Review of: "Advancing Sentiment Analysis: A Novel LSTM Framework with Multi-head Attention"
The paper introduces an LSTM-based sentiment analysis model with multi-head attention and TF-IDF integration. While the proposed approach is conceptually interesting, the presentation can be improved for clarity and precision. Some expressions requir...
- 4 June 2025
- Published by Qeios Ltd
Open AccessReview of: "Advancing Sentiment Analysis: A Novel LSTM Framework with Multi-head Attention"
The paper introduces an LSTM-based sentiment analysis model with multi-head attention and TF-IDF integration. While the proposed approach is conceptually interesting, the presentation can be improved for clarity and precision. Some expressions requir...
- Open Access
- 20 May 2025
- Published by PeerJ in PeerJ Computer Science
- Vol. 11, e2906
Adaptive regularized spectral reduction for stabilizing ill-conditioned bone-conducted speech signals
Bone-conducted (BC) speech signals are inherently challenging to analyze due to their wide frequency range, which leads to ill-conditioning in numerical analysis and linear prediction (LP) techniques. This ill-conditioning is primarily caused by the expansion of eigenvalues, which complicates the stability and accuracy of traditional methods. To address this issue, we propose a novel regularized spectral reduction (RSR) method, built upon the regularized least squares (RLS) framework. The RSR method compresses the frequency range of BC speech signals, effectively reducing eigenvalue spread and enhancing the robustness of LP analysis. Key to the RSR approach is a regularization parameter, fine-tuned iteratively to achieve optimal performance. Experimental results demonstrate that RSR significantly outperforms existing techniques in eigenvalue compression, resulting in more accurate LP analysis for both synthetic and real BC speech datasets. These improvements hold promise for applications in hearing aids, voice recognition systems, and speaker identification in noisy environments, where reliable BC speech analysis is critical.
- 20 May 2025
- Published by PeerJ in PeerJ Computer Science
- Vol. 11, e2906
Open AccessAdaptive regularized spectral reduction for stabilizing ill-conditioned bone-conducted speech signals
Bone-conducted (BC) speech signals are inherently challenging to analyze due to their wide frequency range, which leads to ill-conditioning in numerical analysis and linear prediction (LP) techniques. This ill-conditioning is primarily caused by the expansion of eigenvalues, which complicates the stability and accuracy of traditional methods. To address this issue, we propose a novel regularized spectral reduction (RSR) method, built upon the regularized least squares (RLS) framework. The RSR method compresses the frequency range of BC speech signals, effectively reducing eigenvalue spread and enhancing the robustness of LP analysis. Key to the RSR approach is a regularization parameter, fine-tuned iteratively to achieve optimal performance. Experimental results demonstrate that RSR significantly outperforms existing techniques in eigenvalue compression, resulting in more accurate LP analysis for both synthetic and real BC speech datasets. These improvements hold promise for applications in hearing aids, voice recognition systems, and speaker identification in noisy environments, where reliable BC speech analysis is critical.
- Open Access
- 16 November 2024
- Published by Qeios Ltd
Review of: "Empowering Dysarthric Speech: Leveraging Advanced LLMs for Accurate Speech Correction and Multimodal Emotion Analysis"
DETAILED REVIEW OF THE ARTICLE The paper presents an interesting approach using advanced LLMs (GPT-4, LLaMA) for dysarthric speech correction and emotion recognition. However, it lacks clarity in key areas, such as the validation of pseudo-labels an...
- 16 November 2024
- Published by Qeios Ltd
Open AccessReview of: "Empowering Dysarthric Speech: Leveraging Advanced LLMs for Accurate Speech Correction and Multimodal Emotion Analysis"
DETAILED REVIEW OF THE ARTICLE The paper presents an interesting approach using advanced LLMs (GPT-4, LLaMA) for dysarthric speech correction and emotion recognition. However, it lacks clarity in key areas, such as the validation of pseudo-labels an...
- Open Access
- 3 August 2022
- Published by PeerJ in PeerJ Computer Science
- Vol. 8, e1053
Data augmentation and deep neural networks for the classification of Pakistani racial speakers recognition
Speech emotion recognition (SER) systems have evolved into an important method for recognizing a person in several applications, including e-commerce, everyday interactions, law enforcement, and forensics. The SER system’s efficiency depends on the length of the audio samples used for testing and training. However, the different suggested models successfully obtained relatively high accuracy in this study. Moreover, the degree of SER efficiency is not yet optimum due to the limited database, resulting in overfitting and skewing samples. Therefore, the proposed approach presents a data augmentation method that shifts the pitch, uses multiple window sizes, stretches the time, and adds white noise to the original audio. In addition, a deep model is further evaluated to generate a new paradigm for SER. The data augmentation approach increased the limited amount of data from the Pakistani racial speaker speech dataset in the proposed system. The seven-layer framework was employed to provide the most optimal performance in terms of accuracy compared to other multilayer approaches. The seven-layer method is used in existing works to achieve a very high level of accuracy. The suggested system achieved 97.32% accuracy with a 0.032% loss in the 75%:25% splitting ratio. In addition, more than 500 augmentation data samples were added. Therefore, the proposed approach results show that deep neural networks with data augmentation can enhance the SER performance on the Pakistani racial speech dataset.
- 3 August 2022
- Published by PeerJ in PeerJ Computer Science
- Vol. 8, e1053
Open AccessData augmentation and deep neural networks for the classification of Pakistani racial speakers recognition
Speech emotion recognition (SER) systems have evolved into an important method for recognizing a person in several applications, including e-commerce, everyday interactions, law enforcement, and forensics. The SER system’s efficiency depends on the length of the audio samples used for testing and training. However, the different suggested models successfully obtained relatively high accuracy in this study. Moreover, the degree of SER efficiency is not yet optimum due to the limited database, resulting in overfitting and skewing samples. Therefore, the proposed approach presents a data augmentation method that shifts the pitch, uses multiple window sizes, stretches the time, and adds white noise to the original audio. In addition, a deep model is further evaluated to generate a new paradigm for SER. The data augmentation approach increased the limited amount of data from the Pakistani racial speaker speech dataset in the proposed system. The seven-layer framework was employed to provide the most optimal performance in terms of accuracy compared to other multilayer approaches. The seven-layer method is used in existing works to achieve a very high level of accuracy. The suggested system achieved 97.32% accuracy with a 0.032% loss in the 75%:25% splitting ratio. In addition, more than 500 augmentation data samples were added. Therefore, the proposed approach results show that deep neural networks with data augmentation can enhance the SER performance on the Pakistani racial speech dataset.
- Open Access
- 31 March 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 37149-37163
Recognizing Semi-Natural and Spontaneous Speech Emotions Using Deep Neural Networks
We needed to find deep emotional features to identify emotions from audio signals. Identifying emotions in spontaneous speech is a novel and challenging subject of research. Several convolutional neural network (CNN) models were used to learn deep segment-level auditory representations of augmented Mel spectrograms. The proposed study introduces a novel technique for recognizing semi-natural and spontaneous speech emotions based on 1D (Model A) and 2D (Model B) deep convolutional neural networks (DCNNs) with two layers of long-short-term memory (LSTM). Both models used raw speech data and augmented (mid, left, right, and side) segment level Mel spectrograms to learn local and global features. The architecture of both models consists of five local feature learning blocks (LFLBs), two LSTM layers, and a fully connected layer (FCL). In addition to learning local correlations and extracting hierarchical correlations, LFLB comprises two convolutional layers and a max-pooling layer. The LSTM layer learns long-term correlations from local features. The experiments illustrated that the proposed systems perform better than conventional methods. Model A achieved an average identification accuracy of 94.78% for speaker-dependent (SD) with a raw SAVEE dataset. With the IEMOCAP database, Model A achieved an average accuracy of an SD experiment with raw audio of 73.15%. In addition, Model A obtained identification accuracies of 97.19%, 94.09%, and 53.98% on SAVEE, IEMOCAP, and BAUM-1s, the databases for speaker-dependent (SD) experiments with an augmented Mel spectrogram, respectively. In contrast, Model B achieved identification accuracy of 96.85%, 88.80%, and 48.67% on SAVEE, IEMOCAP, and the BAUM-1s database for SI experiments with augmented reality Mel spectrogram, respectively.
- 31 March 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 37149-37163
Open AccessRecognizing Semi-Natural and Spontaneous Speech Emotions Using Deep Neural Networks
We needed to find deep emotional features to identify emotions from audio signals. Identifying emotions in spontaneous speech is a novel and challenging subject of research. Several convolutional neural network (CNN) models were used to learn deep segment-level auditory representations of augmented Mel spectrograms. The proposed study introduces a novel technique for recognizing semi-natural and spontaneous speech emotions based on 1D (Model A) and 2D (Model B) deep convolutional neural networks (DCNNs) with two layers of long-short-term memory (LSTM). Both models used raw speech data and augmented (mid, left, right, and side) segment level Mel spectrograms to learn local and global features. The architecture of both models consists of five local feature learning blocks (LFLBs), two LSTM layers, and a fully connected layer (FCL). In addition to learning local correlations and extracting hierarchical correlations, LFLB comprises two convolutional layers and a max-pooling layer. The LSTM layer learns long-term correlations from local features. The experiments illustrated that the proposed systems perform better than conventional methods. Model A achieved an average identification accuracy of 94.78% for speaker-dependent (SD) with a raw SAVEE dataset. With the IEMOCAP database, Model A achieved an average accuracy of an SD experiment with raw audio of 73.15%. In addition, Model A obtained identification accuracies of 97.19%, 94.09%, and 53.98% on SAVEE, IEMOCAP, and BAUM-1s, the databases for speaker-dependent (SD) experiments with an augmented Mel spectrogram, respectively. In contrast, Model B achieved identification accuracy of 96.85%, 88.80%, and 48.67% on SAVEE, IEMOCAP, and the BAUM-1s database for SI experiments with augmented reality Mel spectrogram, respectively.
- Open Access
Semi-Natural and Spontaneous Speech Recognition Using Deep Neural Networks with Hybrid Features Unification
Recently, identifying speech emotions in a spontaneous database has been a complex and demanding study area. This research presents an entirely new approach for recognizing semi-natural and spontaneous speech emotions with multiple feature fusion and deep neural networks (DNN). A proposed framework extracts the most discriminative features from hybrid acoustic feature sets. However, these feature sets may contain duplicate and irrelevant information, leading to inadequate emotional identification. Therefore, an support vector machine (SVM) algorithm is utilized to identify the most discriminative audio feature map after obtaining the relevant features learned by the fusion approach. We investigated our approach utilizing the eNTERFACE05 and BAUM-1s benchmark databases and observed a significant identification accuracy of 76% for a speaker-independent experiment with SVM and 59% accuracy with, respectively. Furthermore, experiments on the eNTERFACE05 and BAUM-1s dataset indicate that the suggested framework outperformed current state-of-the-art techniques on the semi-natural and spontaneous datasets.
Open AccessSemi-Natural and Spontaneous Speech Recognition Using Deep Neural Networks with Hybrid Features Unification
Recently, identifying speech emotions in a spontaneous database has been a complex and demanding study area. This research presents an entirely new approach for recognizing semi-natural and spontaneous speech emotions with multiple feature fusion and deep neural networks (DNN). A proposed framework extracts the most discriminative features from hybrid acoustic feature sets. However, these feature sets may contain duplicate and irrelevant information, leading to inadequate emotional identification. Therefore, an support vector machine (SVM) algorithm is utilized to identify the most discriminative audio feature map after obtaining the relevant features learned by the fusion approach. We investigated our approach utilizing the eNTERFACE05 and BAUM-1s benchmark databases and observed a significant identification accuracy of 76% for a speaker-independent experiment with SVM and 59% accuracy with, respectively. Furthermore, experiments on the eNTERFACE05 and BAUM-1s dataset indicate that the suggested framework outperformed current state-of-the-art techniques on the semi-natural and spontaneous datasets.